PG-WM-LATV: a prior-guided weighted multiscale local anisotropic total variation algorithm for missing wedge...

Computed tomography (CT) plays an indispensable role in materials science, biomedical research, and industrial inspection. However, hardware limitations or specimen-related constraints often preclude the acquisition of full-angle projections, leading to the "missing wedge" phenomenon and severely compromising structural fidelity. Local anisotropic total variation (LATV) enhances high-frequency detail preservation by restoring local directional features in the sinogram domain, thereby enabling accurate reconstruction from limited-angle data. Here, we propose a prior-guided weighted multiscale LATV (PG-WM-LATV) algorithm. The algorithm introduces two key constraints in its objective function: a weighted data fidelity term based on the Parker function, and a multi-directional regularization term constructed from prior information. Additionally, we introduce a multiscale post-processing strategy, which employs Gaussian pyramid decomposition to analyze images hierarchically. At each scale, we alternately apply frequency-domain compensation, local projection-domain restoration, and directional TV regularization, achieving an adaptive fusion of global low-frequency structure and local high-frequency detail. We compared the reconstruction performance of the proposed PG-WM-LATV algorithm against total variation (TV), reweighted anisotropic total variation (RwATV), and local anisotropic total variation (LATV) under varying ranges and Poisson noise levels. Both qualitative and quantitative experimental results demonstrate that the PG-WM-LATV algorithm more effectively restores information in regions lost to the missing wedge and significantly enhances structural fidelity in the reconstructed images.